AI pilots at boutique investment banks are most useful when they run on a live mandate, not a dead deal, because a live process creates real edits, deadlines, owner interaction, and review decisions. A dead-deal pilot can show that software loads, but it cannot show how much work remains before a real artifact is sent. Pick an early, ordinary mandate, assign one associate, name the managing director who reviews the work, and let the deal calendar provide the test. This approach also matches general AI assurance guidance, which calls for realistic tests that represent expected use and defined human oversight [1] [2].
What a dead-deal pilot actually tests
When a bank loads a closed or withdrawn mandate into a new tool, it tells itself it is being careful. The data is already clean, the outcome is known, and no client is at risk. All of that is true, and all of it is why the pilot proves little about live production.
The data is clean because an associate spent weeks cleaning it the old way. Feeding a finished data room into a readiness check tells you little about how the tool handles the mess an owner actually sends in week one. The outcome is known, so when the valuation engine produces a figure, the associate compares it to the number the firm already defended and either nods or shrugs. Neither reaction is a useful live-work comparison. And there is no client, so the owner portal, which is the part of the workflow that changes the client's experience most, sits unused.
The pilot also has no deadline. A CIM draft on a dead deal gets marked up when someone has time, which at a boutique bank often means it waits. The question the partner actually cares about, how many hours the draft saves the associate under pressure, cannot be answered without a real production deadline. NIST likewise recommends realistic test sets representative of expected use, not only convenient test conditions [1].
What a dead-deal pilot does test is whether the software loads. That is a demo, and the vendor already gave you one.
Why the edit is the whole experiment
The value of AI in sell-side production sits in one place: the gap between a blank page and a draft that a banker edits. If the draft is good, the edit takes an afternoon. If the draft is bad, the banker rewrites and the tool costs time instead of saving it. Everything else, the valuation run, the buyer list, the pipeline, is measured the same way: how much human work remains after the software produces its output.
That measurement only exists when someone has to produce the finished artifact. On a live mandate, the associate edits the CIM draft because the teaser is going out on the bank's schedule and the CIM has to be ready for the first NDA. The associate notices which sections the tool got right, which it got generic, and which it got wrong, because each one costs real minutes to fix. By the end, the associate can say: the financial summary needed no changes, the market section needed a rewrite, and the thesis needed the partner.
That sentence is the pilot's result. A dead deal cannot produce it. The review is not optional: the OECD AI Principles call for human agency and oversight, with safeguards appropriate to the context [2].
Choosing the right live mandate
Not every live mandate is a good pilot. The wrong choice is the biggest mandate in the building, because the partner on it will not tolerate any learning curve and may pull the associate back to the old workflow at the first friction. The other wrong choice is the stalled mandate, the one where the owner has gone quiet and the partner is half hoping it dies, because that mandate will not generate the deadlines the pilot needs.
The right mandate has three properties. It is early: the engagement letter is signed and the questionnaire has not yet gone out, or has gone out within the last week. It is ordinary: a lower-middle-market company in one of the verticals the bank works in regularly, with an owner who responds to email. And it has a partner who agreed to the pilot in advance and understands that the associate will produce drafts from the tool and then edit them, not rebuild from scratch.
If the firm has two mandates that qualify, pick the one with the more organized owner. The point of the first pilot is to test the tool in expected working conditions, not to test the client.
Running the pilot without slowing the deal
The main objection to a live pilot is risk: what if the tool slows the mandate down or produces something embarrassing. Both are managed the same way every other risk on a mandate is managed, by a named reviewer.
Set the rule on day one. Every artifact the tool produces is a draft until the associate edits it, and every edited draft is a draft until the managing director reads it. Nothing changes about who signs what. The only thing that changes is that the associate starts from a draft instead of a template. That operating rule gives the bank a clear human checkpoint, rather than treating the pilot output as a decision.
Then let the deal's calendar run the pilot. The questionnaire goes to the owner through the portal because that is faster than email, and the associate watches the readiness view fill in. The valuation runs when the financials arrive. The teaser and CIM drafts are generated when the valuation is settled. The buyer list is built from documented criteria when the CIM is finalized, and the senior banker marks it up. NDAs go out from the pipeline. At no point does the pilot add a step; it replaces the blank-page step in each phase while the bank retains control of every buyer contact and decision.
Keep one record: the associate's hours on production work, by phase, in a simple log. That is the only pilot-specific overhead and it takes two minutes a day.
What the pilot should produce by the time NDAs go out
Set the pilot's end at the first NDA out the door, when that is the mandate's actual next decision point. Do not promise a universal timetable. By then, the pilot may have touched discovery and readiness, valuation, marketing materials, and buyer matching, giving the bank enough evidence to decide whether to continue.
The deliverable to the partnership is short. One page from the associate: hours by phase against the associate's own estimate of what the old workflow would have taken, which artifacts needed the least editing and which needed the most, and what the owner said about the portal. One paragraph from the managing director: whether the CIM they signed was better, worse, or the same as the ones they usually sign, and whether they would run the next mandate the same way.
If the answer is yes, the second mandate goes on the platform after the partnership approves the next step. If the answer is no, the bank has a specific reason, tied to a specific artifact on a specific deal, which is the only kind of no worth acting on.
The thing to do this week
Look at the mandates in the building and find the one that was signed most recently. If the questionnaire has not gone out, that mandate is the candidate pilot. Name the associate. Tell the partner. Start the hours log on Monday.
If no mandate qualifies, wait for the next engagement letter rather than loading a dead deal to feel productive. A pilot that waits for a live mandate is delayed. A pilot on a dead deal may need to be run again because it never tested the work the bank actually sends.
How Acquivo Pro fits
Acquivo Pro runs the live mandate on one login, which is what makes a real pilot possible. The associate sends the questionnaire through the branded owner portal, runs a deterministic valuation that traces every figure to its source, generates the teaser and CIM from the questionnaire and data room, builds a buyer list using documented acquisition criteria, and manages NDAs, IOIs, and LOIs in a single pipeline. One seat runs multiple concurrent processes on a flat monthly subscription with no per-deal fees, so the bank can compare the pilot mandate with the next one without changing the workflow. To see it on a live mandate, request a demo through the Contact page.
Key Points
- An AI pilot on a dead deal at a boutique bank tests whether the software loads, not whether it saves associate hours, because nothing produced will be edited or sent under a live deadline.
- The result of an AI pilot at a boutique bank is one sentence from the associate about which artifacts needed no editing, which needed a rewrite, and which needed the partner.
- The right pilot mandate for a boutique bank is early, ordinary, and run by a partner who agreed in advance that the associate will edit drafts rather than rebuild them.
- A live AI pilot at a boutique bank retains human control when every artifact stays a draft until the associate edits it and the managing director reads it.
- A boutique bank should end its first AI pilot at the first NDA out the door and decide on the second mandate from a one-page associate memo and one partner paragraph.
Sources and references
Source retrieval dates record research access, not article publication. Read our editorial and date policy.
Choose your next step
Acquivo is user-operated software. Owners and their professionals remain responsible for decisions, communications, and professional review.
Educational information, not individual legal, tax, or investment advice. Consult qualified professionals for your circumstances. Acquivo provides software, not representation or negotiation on your behalf.